Method and system for predicting main steam temperature of thermal power generating unit

Abstract

The invention discloses a thermal power generating unit main steam temperature prediction method and system, and the method comprises the steps: carrying out the projection of the multivariable time series data of a thermal power generating unit, building space and time node data, carrying out the embedding of the space and time node data, dynamically building an adjacent matrix, and obtaining a heterogeneous space-time diagram; updating the adjacency matrix by utilizing a learnable module based on the interaction relationship of different node types in the adjacency matrix; and fusing different types of neighbor nodes and relationships of the adjacent matrix by using a graph convolution module, updating the heterogeneous space-time graph, inputting the updated heterogeneous space-time graph into the time sequence convolution network, and obtaining a main steam temperature prediction result of the thermal power generating unit. According to the method, the steam related parameters such as the main steam temperature of the thermal power generating unit are efficiently and accurately predicted, the limitation of traditional single-variable time sequence prediction and only concerning the homogeneous relation is exceeded, the complex time-space relation in the multivariable time sequence is comprehensively captured and modeled, and the prediction precision is remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of thermal power units, and in particular to a method and system for predicting the main steam temperature of a thermal power unit, as well as electronic equipment and a storage medium. Background Art

[0002] During the operation of the unit, steam-related control parameters such as initial and final steam pressure, steam temperature and main steam flow are crucial. Keeping these parameters / thresholds within a stable range has become a problem that power plant staff do not need to consider, so efficient and accurate prediction of gas pressure and steam temperature to adjust fuel input has become a new problem that must be solved.

[0003] Existing univariate time series prediction methods are difficult to meet the needs of efficiency and accuracy, and are costly. Therefore, multivariate time series (MVTS) prediction has become a research hotspot. MVTS data contains complex spatial and temporal relationships, including homogeneous and heterogeneous relationships, which need to be fully explored to improve prediction accuracy. Traditional statistical methods are limited by linear assumptions and are difficult to handle complex high-dimensional data. Deep learning methods, especially deep neural networks (DNNs), effectively capture complex dependencies through specially designed network structures. RNNs and their variants are good at modeling temporal relationships ( <t-t>), while convolutional networks and Transformer further enhance the ability to extract temporal features. However, these methods mostly focus on the temporal dimension and ignore the spatial relationship ( <s-s>) diversity. Graph neural networks (GNNs) have attracted much attention due to their ability to flexibly represent relationships between entities. In MVTS prediction, building a graph that accurately reflects the underlying relationships is key. Existing methods mostly rely on static graphs or adaptive matrix learning. <s-s>However, the influence of time on spatial relationships is often ignored. <s-s>relationship, but does not fully consider the heterogeneous relationship in time and space. Summary of the invention

[0004] The purpose of this section is to summarize some aspects of embodiments of the present invention and briefly introduce some preferred embodiments. Some simplifications or omissions may be made in this section and the specification abstract and the invention title of this application to avoid blurring the purpose of this section, the specification abstract and the invention title, and such simplifications or omissions cannot be used to limit the scope of the present invention.

[0005] In view of the above existing problems, the present invention is proposed. Therefore, the present invention provides a method for predicting the main steam temperature of a thermal power unit to solve the problem that the complex spatiotemporal relationship cannot be fully modeled in the multivariate time series prediction in the prior art.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] In a first aspect, the present invention provides a method for predicting main steam temperature of a thermal power unit, comprising:

[0008] Obtain multivariate time series data of thermal power units;

[0009] Projecting the multivariate time series data of the thermal power unit to construct space and time node data, using the space and time node data to embed dynamically construct an adjacency matrix, and obtaining a heterogeneous space-time graph;

[0010] Based on the interaction relationship between different node types in the adjacency matrix, updating the adjacency matrix using a learnable module;

[0011] A graph convolution module is used to fuse different types of neighbor nodes and relationships of the adjacency matrix, the heterogeneous space-time graph is updated, and the updated heterogeneous space-time graph is input into a temporal convolutional network to obtain the main steam temperature prediction result of the thermal power unit.

[0012] As a preferred solution of the main steam temperature prediction method of a thermal power unit described in the present invention, wherein: the interaction relationship based on different node types in the adjacency matrix includes:

[0013] The different node types include time nodes and space nodes;

[0014] The interactive relationship includes time-time relationship, space-space relationship, time-space relationship, space-time relationship;

[0015] A feedforward neural network is used to process the space-time and time-space relationship of the heterogeneous space-time graph, which is expressed as:

[0016]

[0017] Among them, h si and h tj Respectively represent H s The i-th spatial node and H t The jth time node in h tj and h si They represent the i-th time node and the j-th space node respectively.

[0018] As a preferred solution of the method for predicting the main steam temperature of a thermal power unit according to the present invention, wherein: using a learnable module to update the adjacency matrix includes:

[0019] Use the ReLU function to eliminate negative connections in the adjacency matrix;

[0020] Divide the updated adjacency matrix by the maximum value of the adjacency matrix and add a small constant to handle the zero division problem and normalization;

[0021] The tanh function is used to handle the nonlinear dependencies of the adjacency matrix and the edges that are initially zero are kept inactive.

[0022] As a preferred solution of the main steam temperature prediction method of a thermal power unit described in the present invention, wherein: using a graph convolution module to fuse different types of neighbor nodes and relationships of the adjacency matrix includes:

[0023] In the intra-relationship aggregation stage, based on the adjacency matrix, the edge weights between each pair of nodes are calculated to obtain the relative importance of each pair of nodes; the edge weights are normalized between all connected nodes in the same relationship so that the sum of the weights is 1 during aggregation; based on the normalized edge weights, the neighbor information of each node is aggregated for homogeneous and heterogeneous relationships respectively to obtain information representations under different relationships, and the information representations aggregated from different relationships are further fused to construct a comprehensive node representation;

[0024] In the inter-relationship fusion stage, the weights of homogeneous and heterogeneous relationships are calculated; the softmax function is used to normalize all relationships of the same node type to obtain the inter-relationship attention score; based on the attention score, the node representations under different relationships are fused to form an updated node representation.

[0025] As a preferred solution of the method for predicting the main steam temperature of a thermal power unit according to the present invention, the method further comprises:

[0026] A multi-layer heterogeneous space-time graph attention mechanism is used to obtain multi-hop meta-path information of the heterogeneous space-time graph.

[0027] As a preferred solution of the method for predicting the main steam temperature of a thermal power unit described in the present invention, the updated heterogeneous space-time graph is input into a temporal convolutional network to obtain the main steam temperature result of the thermal power unit, including:

[0028] The node representation of the updated heterogeneous space-time graph is reconstructed and superimposed through MLP and GRU, which is expressed as:

[0029]

[0030] Among them, Z s and Z t are the reconstructed spatial and temporal channels, respectively, τ(·) represents the superposition operation, and the parameter C represents the number of information channels;

[0031] The reconstructed and superimposed heterogeneous space-time graph uses a temporal convolutional network to obtain the main steam temperature result of the thermal power unit, which is expressed as:

[0032]

[0033] in, represents the low-level representation extracted by the temporal convolutional network, Represents the 1×1 output convolution kernel, which is used to generate the prediction results of the main steam temperature of the thermal power unit

[0034] As a preferred solution of the method for predicting the main steam temperature of a thermal power unit described in the present invention, the time-time relationship and the space-space relationship include:

[0035] The space-space relationship is represented by the difference or combination of the two adjacency matrices constructed;

[0036] Time-time relations are represented by applying an upper triangular matrix mask to preserve connections from earlier time points to later time points.

[0037] In a second aspect, the present invention provides a system for predicting main steam temperature of a thermal power unit, comprising:

[0038] A data acquisition module is used to obtain multivariate time series data of thermal power units;

[0039] A graph construction module, used for projecting the multivariate time series data of the thermal power unit to construct space and time node data, using the space and time node data to embed dynamically construct an adjacency matrix, and obtaining a heterogeneous space-time graph;

[0040] A graph learning module for updating the adjacency matrix using a learnable module based on the interaction relationship between different node types in the adjacency matrix;

[0041] The prediction module is used to use the graph convolution module to fuse the different types of neighbor nodes and relationships of the adjacency matrix, update the heterogeneous space-time graph, input the updated heterogeneous space-time graph into the temporal convolution network, and obtain the main steam temperature prediction result of the thermal power unit.

[0042] In a third aspect, the present invention provides a computing device, comprising:

[0043] Memory and processor;

[0044] The memory is used to store computer executable instructions, and the processor is used to execute the computer executable instructions. When the computer executable instructions are executed by the processor, the steps of the main steam temperature prediction method of the thermal power unit are implemented.

[0045] In a fourth aspect, the present invention provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the steps of the method for predicting the main steam temperature of a thermal power unit.

[0046] Compared with the prior art, the present invention has the following beneficial effects: By introducing a heterogeneous space-time graph neural network and its unique heterogeneous space-time graph attention mechanism, the present invention achieves efficient and accurate prediction of steam-related parameters such as the main steam temperature of thermal power units, which not only surpasses the limitations of traditional single-variable time series prediction and focusing only on homogeneous relationships, but also comprehensively captures and models the complex spatiotemporal relationships in multivariate time series, significantly improving the prediction accuracy. At the same time, by dynamically optimizing the graph structure and relationship aggregation, the expression ability and generalization ability are further improved, providing strong technical support for power plant operation optimization and fuel input adjustment, reducing operating costs, and improving overall operating efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative labor. Among them:

[0048] Figure 1 It is a schematic diagram of the overall process of a method for predicting the main steam temperature of a thermal power unit according to an embodiment of the present invention;

[0049] Figure 2 A schematic diagram of data acquisition of a method for predicting main steam temperature of a thermal power unit according to an embodiment of the present invention;

[0050] Figure 3 The present invention is a schematic diagram of the overall structure of a method for predicting the main steam temperature of a thermal power unit according to an embodiment of the present invention. DETAILED DESCRIPTION

[0051] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in the art without creative work should fall within the scope of protection of the present invention.

[0052] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0053] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor is it a separate or selective embodiment that is mutually exclusive with other embodiments.

[0054] The present invention is described in detail with reference to schematic diagrams. When describing the embodiments of the present invention, for the sake of convenience, the cross-sectional diagrams showing the device structure will not be partially enlarged according to the general scale, and the schematic diagrams are only examples, which should not limit the scope of protection of the present invention. In addition, in actual production, the three-dimensional dimensions of length, width and depth should be included.

[0055] Meanwhile, in the description of the present invention, it should be noted that the directions or positional relationships indicated by the terms "upper, lower, inner and outer" are based on the directions or positional relationships shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the system or element referred to must have a specific direction, be constructed and operated in a specific direction, and therefore cannot be understood as limiting the present invention. In addition, the terms "first, second or third" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance.

[0056] In the present invention, unless otherwise clearly specified and limited, the terms "install, connect, connect" should be understood in a broad sense, for example: it can be a fixed connection, a detachable connection or an integral connection; it can also be a mechanical connection, an electrical connection or a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0057] Example 1

[0058] Reference Figure 1-3 , is an embodiment of the present invention, and provides a method for predicting the main steam temperature of a thermal power unit, comprising:

[0059] S100: Obtain multivariate time series data of thermal power units;

[0060] In the embodiment of the present application, the multivariate time series data of the thermal power unit includes the data such as steam temperature, steam pressure, steam flow rate, etc. that need to be predicted and collected from the database platform. Among them, 1) the steam temperature data includes the names of the sensor data variables that need to be predicted: main steam temperature 2A, main steam temperature 4A, main steam temperature 5A, reheat steam temperature, main steam temperature before TV1, main steam temperature before TV2, auxiliary steam temperature; 2) the steam pressure data includes the names of the sensor data variables that need to be predicted: cold re-steam pressure, heating extraction steam pressure, auxiliary steam pressure, superheated auxiliary steam pressure, atomizing steam pressure, cold re-steam pressure at the inlet of the reheater, high-pressure steam pressure of desteam pump A, high-pressure steam pressure A, high-pressure steam pressure B, main steam pressure; 3) the steam flow data includes the names of the sensor data variables that need to be predicted: hot re-steam flow, cold re-steam flow, main steam flow, and Flugel calculated main steam flow;

[0061] Specifically, let X = {x1, x2, …, x T }∈R S×T is the multivariate time series data of thermal power units, consisting of S spatial variables and T discrete time steps. and are spatial and temporal datasets, respectively, with X s and X t to represent these spatial and temporal features.

[0062] S102: Projecting the multivariate time series data of the thermal power unit to construct spatial and time node data, using the spatial and time node data to embed dynamically construct an adjacency matrix, and obtaining a heterogeneous space-time graph;

[0063] In the embodiment of the present application, from the X in the input window X s and X t Construct space and time nodes, due to X s and X t The lengths of X s and X t Project to and Where h represents the size of the latent space dimension; Figure 2 As shown;

[0064] Specifically, the projection process is expressed as:

[0065] H s =MLP(X s ),H t =GRU(X t ).

[0066] Among them, H s The spatial node after projection, H t is the time node after projection, X s is the spatial node dataset, X t is the time node dataset;

[0067] In the embodiment of the present application, the spatial node is constructed using a multi-layer perceptron (MLP), and the temporal node is constructed using a gated recurrent unit (GRU) for extracting temporal features;

[0068] In the embodiment of the present application, unlike using a static graph, the present invention uses node embedding to dynamically construct the adjacency matrix from updated node embeddings. The adjacency matrix changes over time. The process of constructing the adjacency matrix is ​​expressed as:

[0069]

[0070] Among them, A Ψ is the adjacency matrix, H Φ1 and H Φ2 are the representations of the source node and the target node respectively;

[0071] It should be noted that the present invention cleverly projects the complex and changeable time series data of thermal power units to potential space and time dimensions, which not only effectively integrates the static characteristics of spatial nodes and the dynamic change characteristics of time nodes, but also uses multi-layer perceptrons and gated recurrent units to enhance the ability to extract spatial information and capture time series features respectively; further, by dynamically constructing an adjacency matrix, real-time updating of heterogeneous space-time graphs is achieved. The present invention not only reflects the complex and changeable dependencies between data, but also enables the graph structure to flexibly adapt to the dynamic changes of the system state, thereby improving the analysis accuracy and real-time performance of the thermal power unit operation data, and providing a more solid data support and model foundation for subsequent fault diagnosis, performance optimization and intelligent control.

[0072] S104: Based on the interaction relationship between different node types in the adjacency matrix, the adjacency matrix is ​​updated using a learnable module;

[0073] Preferably, the different node types include time nodes and space nodes;

[0074] Preferably, the interactive relationship includes a time-time relationship, a space-space relationship, a time-space relationship, and a space-time relationship;

[0075] Preferably, a feed-forward neural network is used to process the space-time and time-space relationships of the heterogeneous space-time graph, expressed as:

[0076]

[0077] Among them, h si and h tj Respectively represent H s The i-th spatial node and H t The jth time node in h tj and h si Represent the i-th time node and the j-th space node respectively;

[0078] Preferably, the space-space relationship is represented by the difference or combination of two constructed adjacency matrices;

[0079] Specifically, in the multivariate time series data of thermal power units, the spatial node V s The interaction between them should be unidirectional. The present invention constructs two matrices M1 and M2 based on the learned representation and uses subtraction to obtain the unidirectional influence between spatial nodes. The calculation process is expressed as:

[0080]

[0081] Among them, H s The spatial node after projection, H t is the time node after projection, Θ1 and Θ2 are parameters, α is a hyperparameter that controls the saturation rate of the activation function, and G ss for space-space relations;

[0082] Preferably, the time-time relationship is represented by applying an upper triangular matrix mask to preserve connections from earlier time points to later time points;

[0083] Specifically, the time-time relationship reflects the flow between time factors, so it is reasonable to have only connections from nodes at earlier time points to nodes at later time points. In order to ensure the correct direction flow in the time domain, the present invention applies a mask to remove inappropriate connections, which is expressed as:

[0084]

[0085] Among them, M tt is the time-time relationship, H t is the time node after projection, Triu(·) represents the process of masking the triangular part under the matrix product, and the self-connection is still preserved in this operation;

[0086] Preferably, the ReLU function is used to eliminate negative connections in the adjacency matrix;

[0087] Divide the updated adjacency matrix by the maximum value of the adjacency matrix and add a small constant to handle the zero division problem and normalization;

[0088] Preferably, the tanh function is used to process the nonlinear dependency of the adjacency matrix, and the edges that are initially zero are kept inactive;

[0089] In the embodiment of the present application, in order to further optimize the graph structure by considering the interaction between nodes, the present invention uses a learnable module to learn the interaction and update the initial adjacency matrix A of each relationship. Ψ , the calculation process is as follows:

[0090] M Ψ =ReLU(G Ψ (H Φ1 ,H Φ2 )+A Ψ ),

[0091]

[0092] Among them, G Ψ is a learnable module, M Ψ is the updated matrix, and the ReLU(·) function is used to eliminate negative connections;

[0093] In order to normalize the matrix while keeping the edges removed by the ReLU function inactive, M Ψ Divide by its maximum value and add a small constant to avoid division by zero, use the tanh(·) function to find nonlinear dependencies, and keep the non-existent edges, i.e., M Ψ The zero value in is specifically expressed as:

[0094]

[0095] Among them, M Ψ is the updated matrix, tanh(·) is the function for handling nonlinear relationships, and σ is a small constant;

[0096] It should be noted that, through the above method, the present invention effectively utilizes learnable modules to dynamically update the adjacency matrix in the heterogeneous space-time graph, fully considering the complex interaction relationship between different node types; by processing the space-time and time-space relationships through a feedforward neural network, combined with specific matrix operations and activation functions (such as ReLU and tanh), not only the unreasonable negative connections are eliminated, but also the important unidirectional influence and time flow directionality are retained, while the nonlinear expression ability and numerical stability of the matrix are enhanced. This adaptive graph structure update mechanism provides a more accurate and flexible graph representation for processing complex multivariate time series data, significantly improving the performance and effect when processing such data.

[0097] S106: Use the graph convolution module to fuse different types of neighbor nodes and relationships of the adjacency matrix, update the heterogeneous space-time graph, input the updated heterogeneous space-time graph into the temporal convolution network, and obtain the main steam temperature prediction result of the thermal power unit;

[0098] Preferably, in the intra-relation aggregation stage, the edge weights between each pair of nodes are calculated based on the adjacency matrix to obtain the relative importance of each pair of nodes; the edge weights are normalized between all connected nodes in the same relationship so that the sum of the weights is 1 during aggregation; based on the normalized edge weights, the neighbor information of each node is aggregated for homogeneous and heterogeneous relationships respectively to obtain information representations under different relationships, and the information representations aggregated in different relationships are further fused to construct a comprehensive node representation;

[0099] In the embodiment of the present application, intra-relation aggregation aims to aggregate information from source nodes to target nodes. Each node has two relationships: a homogeneous relationship and a heterogeneous relationship, such as the relationship involving spatial nodes. <s-s>and <t-s>, homogeneous / heterogeneous relations have different semantics, so each relationship should be handled separately;

[0100] In the aggregation process of each relationship Ψ, the present invention only considers E Ψ In order to utilize and promote the learning of the adjacency matrix, edge weight attention is used to incorporate edge weights in aggregation;

[0101] Edge-weighted attention Edge-weighted attention first calculates the edge importance of each pair of nodes (i, j) using the edge weights obtained from the learned adjacency matrix and node features. The calculation process is expressed as:

[0102]

[0103] in represents the edge weight between source node i and target node j, a Ψ is a parameter vector specific to the relation Ψ, e ij For edge importance;

[0104] The edge importance e ij Normalize across all nodes connected in the same relationship:

[0105]

[0106] Then, information is aggregated between neighbors of different relationships. The aggregation process for node j is expressed as:

[0107]

[0108] in is the representation of node j in relation Ψ, h i is the characteristic of source node i;

[0109] Repeat the polymerization process to obtain and T t E The representation of each relation in ss ,O ts ) and (O tt ,O st ), these representations are further fused into new spatial and temporal representations;

[0110] Preferably, in the inter-relationship fusion stage, the weights of homogeneous and heterogeneous relationships are calculated; the softmax function is used to normalize all relationships of the same node type to obtain the attention scores between the relationships; according to the attention scores, the node representations under different relationships are fused to form an updated node representation;

[0111] In the embodiment of the present application, the relationship fusion combines the homogeneous and heterogeneous relationship representations of each node, for example and Fusion to spatial node i, similarly, the relationship importance is first calculated as:

[0112]

[0113] Among them, O Ψ The representation of the relationship Ψ, q Ψ , W Ψ and b Ψ is a learnable parameter;

[0114] All relations of target node type Φ through softmax function Normalized relationship importance w Ψ The attention score between relations is obtained, expressed as:

[0115]

[0116] where β Ψ is the attention score of relation Ψ;

[0117] Fuse the relationship representations to obtain updated node representations:

[0118]

[0119] Among them, H′ Φ is the updated node representation of node type Φ;

[0120] For all Φ∈T V Repeat this process to obtain the updated spatial and temporal node representation H′ s and H′ t ;

[0121] Preferably, a multi-layer heterogeneous space-time graph attention mechanism is used to obtain multi-hop meta-path information of the heterogeneous space-time graph;

[0122] Specifically, a single layer of the heterogeneous spatial-temporal graph attention mechanism focuses on extracting information from nodes with one-hop relations;

[0123] However, in the multivariate time series data of thermal power plants, there may be meta-path-based information, e.g. <s-s-s>and <s-t-t>To mine this information, we use K layers of heterogeneous spatial-temporal graph attention mechanisms to explore insights in K-hop meta-paths and output updated spatial and temporal node representations H s ′ and H t ′.

[0124] Preferably, the node representation of the updated heterogeneous space-time graph is reconstructed and superimposed through MLP and GRU, which is expressed as:

[0125]

[0126] Among them, Z s and Z t are the reconstructed spatial and temporal channels, respectively, τ(·) represents the superposition operation, and the parameter C represents the number of information channels;

[0127] In the embodiment of the present application, after reconstruction, a 1×1 starting convolution is used to map the input channel to multiple channels;

[0128] Then comes the dilated inception layer. The dilated inception layer consists of dilated convolutions with different kernels. The dilated convolution is defined as follows:

[0129]

[0130] where d is the dilation factor, f 1×m represents a 1D convolution kernel, Represents a convolution operation. The expanded inception layer contains a set of 1D filters with different kernel sizes, 1×m1, 1×m2, …, 1×m ω , used to extract multiple patterns, for example, a kernel size of 1×7 can be used to capture weekly periodicity;

[0131] The outputs of different filters are truncated to the same length according to the maximum filter size and concatenated in the channel dimension. The calculation of the expanded inception layer is as follows:

[0132]

[0133] in, Represents a set of convolution kernels with different sizes. The Trunc operation refers to truncating the length of these convolution outputs to the length determined by the maximum kernel size. The Concat operation represents the concatenation of the results along the channel dimension.

[0134] In order to control the flow of information in the expanded inception layer, a gate mechanism is used on the two expanded inception layers;

[0135] Specifically, two different activation functions are used to control the amount of information passed by the filter to the next layer, and then the results are multiplied element by element, as follows:

[0136] Q = σ(Q′ f )⊙μ(Q′ g ),

[0137] Where σ represents the sigmoid function, μ represents the hyperbolic tangent function, ⊙ is the Hadamard product, Q′ f and Q g ′ represents the output of the filter and gated dilation inception respectively;

[0138] To expand the receptive field and handle longer dependencies, the dilation factor grows exponentially in each layer of the temporal convolutional module; in addition, to train deeper networks, residual connections are used in each layer to pass the input directly to the next layer.

[0139] Preferably, the reconstructed and superimposed heterogeneous space-time graph uses a temporal convolutional network to obtain the main steam temperature result of the thermal power unit, which is expressed as:

[0140]

[0141] in, represents the low-level representation extracted by the temporal convolutional network, Represents the 1×1 output convolution kernel, which is used to generate the prediction results of the main steam temperature of the thermal power unit

[0142] It should be noted that, through the above detailed steps, the present invention not only realizes the deep fusion and update of nodes and relationships in heterogeneous space-time graphs, but also effectively extracts multi-level information in the graph structure through intra-relationship aggregation and inter-relationship fusion strategies; in particular, the multi-layer heterogeneous space-time graph attention mechanism is used to explore the potential information in multi-hop meta-paths, which significantly enhances the model's ability to understand complex data relationships;

[0143] It should also be noted that the present invention not only expands the receptive field of the model, but also improves its efficiency and accuracy in processing long-term dependencies by reconstructing node representation and cleverly combining MLP, GRU and expanded inception layers. Finally, combined with the temporal convolutional network, the present invention can accurately predict the main steam temperature of thermal power units, providing strong technical support for industrial process optimization and control.

[0144] The above is a schematic scheme of a method for predicting the main steam temperature of a thermal power unit in this embodiment. It should be noted that the technical scheme of the system for predicting the main steam temperature of a thermal power unit and the technical scheme of the method for predicting the main steam temperature of a thermal power unit described above are of the same concept. For details not described in detail in the technical scheme of the system for predicting the main steam temperature of a thermal power unit in this embodiment, please refer to the description of the technical scheme of the method for predicting the main steam temperature of a thermal power unit described above.

[0145] The main steam temperature prediction system of a thermal power unit in this embodiment includes:

[0146] A data acquisition module is used to obtain multivariate time series data of thermal power units;

[0147] A graph construction module is used to project the multivariate time series data of thermal power units to construct spatial and time node data, and use the spatial and time node data to embed dynamically construct an adjacency matrix to obtain a heterogeneous space-time graph;

[0148] A graph learning module is used to update the adjacency matrix using a learnable module based on the interaction relationship between different node types in the adjacency matrix;

[0149] The prediction module is used to use the graph convolution module to fuse different types of neighbor nodes and relationships in the adjacency matrix, update the heterogeneous space-time graph, input the updated heterogeneous space-time graph into the temporal convolution network, and obtain the main steam temperature prediction result of the thermal power unit.

[0150] This embodiment further provides a computing device, which is applicable to the case of main steam temperature prediction of a thermal power unit, and includes:

[0151] Memory and processor; the memory is used to store computer executable instructions, and the processor is used to execute computer executable instructions to implement the main steam temperature prediction method of a thermal power unit as proposed in the above embodiment.

[0152] This embodiment further provides a storage medium on which a computer program is stored. When the program is executed by a processor, the method for predicting the main steam temperature of a thermal power unit proposed in the above embodiment is implemented.

[0153] The storage medium proposed in this embodiment and the method for realizing the main steam temperature prediction of a thermal power unit proposed in the above embodiment belong to the same inventive concept. The technical details not fully described in this embodiment can be referred to the above embodiment, and this embodiment has the same beneficial effects as the above embodiment.

[0154] Through the above description of the implementation methods, the technicians in the relevant field can clearly understand that the present invention can be implemented by means of software and necessary general hardware, and of course can also be implemented by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ReadOnly, Memory, ROM), random access memory (RandomAccess Memory, RAM), flash memory (FLASH), hard disk or optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, server, or network device, etc.) to perform the methods of various embodiments of the present invention.

[0155] Example 2

[0156] Referring to Table 1, an embodiment of the present invention provides a method for predicting the main steam temperature of a thermal power unit. In order to verify its beneficial effects, comparison results of several schemes are provided.

[0157] In order to verify the effectiveness of the main steam temperature prediction method of thermal power units of the present invention, a series of experiments were designed and implemented;

[0158] Multivariate time series data were collected from the actual operating data of thermal power units. These data include multiple key parameters such as air pressure, steam temperature, steam flow rate, etc. These data are real-time records of thermal power units under different operating conditions, providing rich training and test samples for the prediction model.

[0159] Next, the data is processed according to the steps of the invention, firstly the multivariate time series data is projected into the two dimensions of space and time, the spatial node and time node data are constructed, and the adjacency matrix is ​​dynamically constructed using these data to form a heterogeneous space-time graph;

[0160] After constructing the heterogeneous space-time graph, the adjacency matrix is ​​updated using a learnable module. In this process, the ReLU function is used to eliminate unreasonable negative connections, and normalization and tanh function processing are used to enhance the nonlinear expression ability and numerical stability of the matrix. In this way, a more accurate and adaptable graph structure is obtained.

[0161] Subsequently, the graph convolution module is used to perform intra-relation aggregation and inter-relation fusion on the updated adjacency matrix. In the intra-relation aggregation stage, the edge weights between each pair of nodes are calculated, and the neighbor information of the nodes is aggregated based on these weights to construct a comprehensive node representation.

[0162] In the inter-relationship fusion stage, the softmax function is used to calculate the weights of homogeneous and heterogeneous relations, and the node representations under different relations are fused to form an updated node representation.

[0163] In order to further mine the multi-hop meta-path information in the graph structure, a multi-layer heterogeneous spatial-temporal graph attention mechanism is introduced, which can explore more complex dependencies between nodes and improve the predictive ability of the model.

[0164] Finally, the updated heterogeneous space-time graph is input into the temporal convolutional network. Through the superposition of MLP and GRU and the temporal convolution operation, the low-level representation of the data is extracted and the prediction result of the main steam temperature of the thermal power unit is generated.

[0165] In order to evaluate the performance of this method, the prediction results are compared with a variety of existing multivariate prediction methods, including MTGNN, NLinear, DLinear, FiLM, TSMixer and Crossformer, and the results are shown in Table 1;

[0166] Table 1 Comparison of the method with multivariable prediction results in pressure, steam temperature and steam flow

[0167]

[0168] The results in Table 1 show that the method exhibits lower MSE and MAE values ​​in the prediction of gas pressure, steam temperature and steam flow, especially in the prediction of steam temperature. This result shows that the present invention can more effectively capture the complex relationship of multivariate time series data of thermal power units, improve the accuracy and stability of the prediction model, and provide strong support for the optimized operation of thermal power units.

[0169] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A method for predicting main steam temperature of a thermal power unit, characterized in that: include: Obtain multivariate time series data of thermal power units; Projecting the multivariate time series data of the thermal power unit to construct space and time node data, using the space and time node data to embed dynamically construct an adjacency matrix, and obtaining a heterogeneous space-time graph; Based on the interaction relationship between different node types in the adjacency matrix, updating the adjacency matrix using a learnable module; A graph convolution module is used to fuse different types of neighbor nodes and relationships of the adjacency matrix, the heterogeneous space-time graph is updated, and the updated heterogeneous space-time graph is input into a temporal convolutional network to obtain the main steam temperature prediction result of the thermal power unit.

2. The method for predicting the main steam temperature of a thermal power unit according to claim 1, characterized in that: The interactions based on different node types in the adjacency matrix include: The different node types include time nodes and space nodes; The interactive relationship includes time-time relationship, space-space relationship, time-space relationship, space-time relationship; A feedforward neural network is used to process the space-time and time-space relationship of the heterogeneous space-time graph, which is expressed as: Among them, h si and h tj Respectively represent H s The i-th spatial node and H t The jth time node in h tj and h si They represent the i-th time node and the j-th space node respectively.

3. The method for predicting the main steam temperature of a thermal power unit according to claim 1 or 2, characterized in that: Updating the adjacency matrix using a learnable module includes, Use the ReLU function to eliminate negative connections in the adjacency matrix; Divide the updated adjacency matrix by the maximum value of the adjacency matrix and add a small constant to handle the zero division problem and normalization; The tanh function is used to handle the nonlinear dependencies of the adjacency matrix and the edges that are initially zero are kept inactive.

4. The method for predicting the main steam temperature of a thermal power unit according to claim 3, characterized in that: Use the graph convolution module to fuse the adjacency matrix of different types of neighbor nodes and relationships including: In the intra-relationship aggregation phase, the edge weights between each pair of nodes are calculated based on the adjacency matrix to obtain the relative importance of each pair of nodes; Normalize the edge weights between all connected nodes in the same relationship so that the sum of the weights is 1 when aggregated; Based on the normalized edge weights, the neighbor information of each node is aggregated for homogeneous and heterogeneous relationships respectively to obtain information representations under different relationships. The information representations aggregated from different relationships are further fused to construct a comprehensive node representation. In the inter-relationship fusion stage, the weights of homogeneous and heterogeneous relationships are calculated; Use the softmax function to normalize all relations of the same node type and obtain the attention score between relations; According to the attention score, the node representations under different relationships are fused to form an updated node representation.

5. The method for predicting the main steam temperature of a thermal power unit according to claim 4, characterized in that: Also includes, A multi-layer heterogeneous space-time graph attention mechanism is used to obtain multi-hop meta-path information of the heterogeneous space-time graph.

6. The method for predicting the main steam temperature of a thermal power unit according to claim 4, characterized in that: The updated heterogeneous space-time graph is input into the temporal convolutional network to obtain the main steam temperature results of the thermal power unit, including: The node representation of the updated heterogeneous space-time graph is reconstructed and superimposed through MLP and GRU, which is expressed as: Among them, Z s and Z t are the reconstructed spatial and temporal channels, respectively, τ(·) represents the superposition operation, and the parameter C represents the number of information channels; The reconstructed and superimposed heterogeneous space-time graph uses a temporal convolutional network to obtain the main steam temperature result of the thermal power unit, which is expressed as: in, represents the low-level representation extracted by the temporal convolutional network, Represents the 1×1 output convolution kernel, which is used to generate the prediction results of the main steam temperature of the thermal power unit 7. The method for predicting the main steam temperature of a thermal power unit according to claim 2, characterized in that: Time-time relationship, space-space relationship includes, The space-space relationship is represented by the difference or combination of the two adjacency matrices constructed; Time-time relations are represented by applying an upper triangular matrix mask to preserve connections from earlier time points to later time points.

8. A system for predicting the main steam temperature of a thermal power unit, characterized in that: include, A data acquisition module is used to obtain multivariate time series data of thermal power units; A graph construction module, used for projecting the multivariate time series data of the thermal power unit to construct space and time node data, using the space and time node data to embed dynamically construct an adjacency matrix, and obtaining a heterogeneous space-time graph; A graph learning module for updating the adjacency matrix using a learnable module based on the interaction relationship between different node types in the adjacency matrix; The prediction module is used to use the graph convolution module to fuse the different types of neighbor nodes and relationships of the adjacency matrix, update the heterogeneous space-time graph, input the updated heterogeneous space-time graph into the temporal convolution network, and obtain the main steam temperature prediction result of the thermal power unit.

9. An electronic device, comprising: Memory and processor; The memory is used to store computer executable instructions, and the processor is used to execute the computer executable instructions. When the computer executable instructions are executed by the processor, the steps of the main steam temperature prediction method of a thermal power unit described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the steps of the method for predicting the main steam temperature of a thermal power unit as claimed in any one of claims 1 to 7.

Images